Photolithographic mask fabrication for ultra large scale integrated (ULSI) circuits requires a much higher degree of etch uniformity than semiconductor wafer processing. A single mask pattern generally occupies a four inch square area on a quartz mask. The image of the mask pattern is focused down to the area of a single die (a one inch square) on the wafer and is then stepped across the wafer, forming a single image for each die. Prior to etching the mask pattern into the quartz mask, the mask pattern is written by a scanning electron beam, a time consuming process which renders the cost of a single mask extremely high. The mask etch process is not uniform across the surface of the mask. Moreover, the e-beam written photoresist pattern is itself non-uniform, and exhibits, in the case of 45 nm feature sizes on the wafer, as much as 2-3 nm variation in critical dimension (e.g., line width) across the entire mask. (This variation is the 3σ variance of all measured line widths, for example.) Such non-uniformities in photoresist critical dimension will vary among different mask sources or customers. The mask etch process cannot increase this variation by more than 1 nm, so that the variation in the etched mask pattern cannot exceed 3-4 nm. These stringent requirements arise from the use of diffraction effects in the quartz mask pattern to achieve sharp images on the wafer. It is difficult to meet such requirements with current technology. It will be even more difficult for future technologies, which may involve 22 nm wafer feature sizes. This difficulty is compounded by the phenomenon of etch bias, in which the depletion of the photoresist pattern during mask etch causes a reduction in line width (critical dimension) in the etched pattern on the quartz mask. These difficulties are inherent in the mask etch process because the etch selectivity of typical mask materials (e.g., quartz, chrome, molybdenum silicide) relative to photoresist is typically less than one, so that the mask photoresist pattern is etched during the mask etch process.
Some mask patterns require etching periodic openings into the quartz mask by a precisely defined depth that is critical to achieving the extremely fine phase alignment of interfering light beams during exposure of the wafer through the mask. For example, in one type of phase shift mask, each line is defined by a chrome line with thin quartz lines exposed on each side of the chrome line, the quartz line on one side only being etched to a precise depth that provides a 180 degree phase shift of the light relative to light passing through the un-etched quartz line. In order to precisely control the etch depth in the quartz, the etch process must be closely monitored by periodically interrupting it to measure the etch depth in the quartz. Each such inspection requires removing the mask from the mask etch reactor chamber, removing the photoresist, measuring the etch depth and then estimating the etch process time remaining to reach the target depth based upon the elapsed etch process time, depositing new photoresist, e-beam writing the mask pattern on the resist, re-introducing the mask into the mask etch chamber and restarting the etch process. The estimate of remaining etch time to reach the desired depth assumes that the etch rate remains stable and uniform, and therefore is unreliable. The problems of such a cumbersome procedure include low productivity and high cost as well as increased opportunity for contamination or faults in the photoresist pattern. However, because of the requirement for an accurately controlled etch depth, there has seemed to be no way around such problems.
The small tolerance in critical dimension variation requires extremely uniform distribution of etch rate over the mask surface. In masks requiring precise etch depth in the quartz material, there are two critical dimensions, one being the line width and the other being the etch depth, and uniformity for both types of critical dimension requiring a uniform etch rate distribution across the mask. Non-uniformity in etch rate distribution can be reduced to some extent by employing a source power applicator that can vary the radial distribution of the plasma ion density, such as an inductive source power applicator consisting of inner and outer coil antennas overlying the wafer. Such an approach, however, can only address non-uniformities that are symmetrical, that is a center-high or a center-low etch rate distribution. In practice, non-uniformities in etch rate distribution can be non-symmetrical, such as a high etch rate in one corner of the mask, for example. A more fundamental limitation is that the mask etch process tends to have such an extremely center-low distribution of etch rate that a tunable feature, such an inductive power applicator having inner and outer coils, is incapable of transforming the etch rate distribution out of the center-low regime.
Another problem with non-uniform etch rate distribution is that the etch rate distribution tends to vary widely among different reactors of the same design and can vary widely within the same reactor whenever a key part or a consumable component is replaced, such as replacement of the cathode. The etch rate distribution appears to be highly sensitive to small variations in features of the replaced part, with unpredictable changes upon consumable replacement.